I made a small artificial to roam around on my website dmewes.com while the browser is open.
His name is Hebbie, and he can roam around and eat the words on the page. Lower-case words reduce hunger, while capitalized words act as a mild poison.
He learns which actions to take through online Hebbian learning. In particular, he learns small world models about how different actions affect his drives in certain situations, and then uses those models to select the most advantageous action to take.
It’s very simple of course, but enough to generate some mildly non-trivial behaviors and learn basic associations within a few minutes.
There’s also a brain monitor where you can check what’s going on and see it learn.
I love this! I like it when complex subject matter is explored in simplified and playful ways — makes it way easier for beginners to understand. Haven’t found the control to activate the monitor so far, though. (Edit: found it: brain icon in bottom right corner). Still, from the screenshot it looks very clear and useful for learners.
I also see that you have a history with Creatures add-ons and related research. I am wondering if you have any subjective take on the Grandroids architecture, particularly the interplay of maps. I’m on a journey trying to take it all in (starting with the discussions here, but backtracking to old logs and presentations, way back to both books and Creatures paper (I also want to understand a bit how the ideas came to be)), but my picture of the current approach is not quite complete yet. 🙂
